c2023-models-tested-gpt2-neox-llama-gpt3

IN premise — summaries/2026/08/24/cohen-2023-ripple-effects-sR-references.md

Created 2026-08-25T02:57:58+00:00

Cohen et al. (2023) evaluated KE methods across four model architectures: GPT-2, GPT-NeoX, LLAMA, and GPT-3

Summary

The 2023 Cohen study validated its knowledge-engineering approach only against four specific transformer model families (GPT-2, GPT-NeoX, LLAMA, GPT-3), so its findings carry weight for those architectures but don't automatically extend to models outside that set. This acts as a scope boundary: any conclusions in the system that rely on that study are only as broad as those four model families.